A Fixed Claim on a Shrinking Supply: AI Data Centres, Water Stress and Lock-In

Public discussion increasingly presents artificial intelligence as a technology that takes water from people. This paper tests that claim against the scale of human freshwater use and examines where AI infrastructure is being built. Data centres withdrew about 5.5 km³ of fresh water in 2023, roughly 0.14% of human withdrawals; livestock production consumes about 270 times as much surface and groundwater, and leaking supply networks lose about 23 times what data centres withdraw. At global scale, AI is not a cause of freshwater scarcity. The risk is local and lies in siting: about half of the world’s data-centre hubs, and two-thirds of new US data centres since 2022, are in high water-stress areas. The paper proposes three tracks: non-potable cooling water and low-water electricity in water-stressed basins; siting assessments over each facility’s operating life; and AI applied to reducing losses in local water networks.

Document viewer may not display on some mobile browsers — use the download link above.